nnterp · encyclopedia youtu · nnterp.families.youtu

family youtu

Youtu-LLM

Llama's dense block with DeepSeek-V2's latent attention: queries and keys wider than values, and every head's key and value expanded from one latent per token.

Latent attentionInterleaved rotary

The block

01 · the residual stream

One block of the stack. Hover any part for its nnterp name; click to pin the card.

embed_tokenstoken_embeddings
→
layers× 32
→
normfinal_norm_output
→
lm_headlm_head_output
→
logitsthe output
i = 0

identity layers[i].input + self_attn.attention_output + mlp.mlp_output == layer_output

The API

02 · support()

Values, by host

⚠︎ marks a value this checkpoint has only under some load: hover it for the condition.

model YoutuForCausalLMlayoutwhat it isread at
logits Logits [batch seq vocab_size] The logits the model returns, after anything it does past lm_head (a softcap, a scale)

the module's own output
token_embeddings Residual [batch seq hidden_size] The embedding module's output; layers[0].input is what enters the first block

the child module embed_tokens's output
final_norm_output Residual [batch seq hidden_size] The residual stream after the final norm, what lm_head reads

the child module norm's output
lm_head_output Logits [batch seq vocab_size] The head's logits before anything the model does past lm_head (a softcap, a scale)

the child module lm_head's output
next_token_probs NextTokenProbs [batch vocab_size] The next-token distribution at the last position; derived, read-only

the module's own output
input_ids Tokens [batch seq] The token ids the model was called with

the module's own input
attention_mask Tokens [batch seq] The attention mask the model was called with; zeros are padding

the module's own input
input_size scalar [batch, seq] of the current call; read-only

the module's own input
model.layers[i] YoutuDecoderLayerlayoutwhat it isread at
layer_input Residual [batch seq hidden_size] The residual stream entering the block

the module's own input
attention_norm_output Residual [batch seq hidden_size] The normed stream entering the attention

the child module input_layernorm's output
layer_mid Residual [batch seq hidden_size] The residual stream after the attention add

the child module post_attention_layernorm's input
mlp_norm_output Residual [batch seq hidden_size] The normed stream entering the MLP

the child module post_attention_layernorm's output
mlp_input Residual [batch seq hidden_size] What the MLP is called with

the child module mlp's input
layer_output Residual [batch seq hidden_size] The residual stream leaving the block, a tensor even when the block returns a tuple

the module's own output
model.layers[i].self_attn YoutuAttentionlayoutwhat it isread at
⚠︎attention_queries Queries [batch num_heads seq qk_head_dim] The queries entering attention

inside the forward: the arguments of attention_interface_1, element 1
⚠︎attention_keys Keys [batch num_kv_heads seq qk_head_dim] The keys entering attention

inside the forward: the arguments of attention_interface_1, element 2
⚠︎attention_values Values [batch num_kv_heads seq head_dim] The values entering attention

inside the forward: the arguments of attention_interface_1, element 3
⚠︎attention_scores Pattern [batch num_heads query key] The attention scores entering the softmax, masked

inside the forward: the input of nn_functional_softmax_0 inside attention_interface_1
attention_premix HeadOutputs [batch seq num_heads head_dim] Every head's output side by side, as the output projection takes it

the child module o_proj's input
attention_output Residual [batch seq hidden_size] What the attention adds to the residual stream

the module's own output
⚠︎attention_probabilities Pattern [batch num_heads query key] The attention pattern the values are mixed with

inside the forward: the output of nn_functional_dropout_0 inside attention_interface_1
⚠︎attention_head_outputs HeadOutputs [batch seq num_heads head_dim] The per-head outputs before the output projection

inside the forward: the output of attention_interface_1, element 0
model.layers[i].mlp YoutuMLPlayoutwhat it isread at
mlp_activation Neurons [batch seq intermediate_size] The activation function's output

the child module act_fn's output
mlp_neurons Neurons [batch seq intermediate_size] What the down projection receives: act(gate)·up on a gated MLP, the activation itself otherwise

the child module down_proj's input
mlp_output Residual [batch seq hidden_size] What the MLP adds to the residual stream

the module's own output

Printout

What print(model) shows on tencent/Youtu-LLM-2B-Base: the native modules, standard names and every standard value with its layout.

YoutuForCausalLM(
  (lm_head): Linear(in_features=2048, out_features=128256, bias=False)
  (generator): Generator(
    (streamer): Streamer()
  )
  (embed_tokens): Embedding(128256, 2048)
  (layers): ModuleList(
    (0-31): 32 x YoutuDecoderLayer(
      (self_attn): YoutuAttention(
        (q_a_proj): Linear(in_features=2048, out_features=1536, bias=False)
        (q_a_layernorm): YoutuRMSNorm((1536,), eps=1e-06)
        (q_b_proj): Linear(in_features=1536, out_features=3072, bias=False)
        (kv_a_proj_with_mqa): Linear(in_features=2048, out_features=576, bias=False)
        (kv_a_layernorm): YoutuRMSNorm((512,), eps=1e-06)
        (kv_b_proj): Linear(in_features=512, out_features=4096, bias=False)
        (o_proj): Linear(in_features=2048, out_features=2048, bias=False)
        (attention_queries) -> Queries [batch num_heads seq qk_head_dim]: The queries entering attention
        (attention_keys) -> Keys [batch num_kv_heads seq qk_head_dim]: The keys entering attention
        (attention_values) -> Values [batch num_kv_heads seq head_dim]: The values entering attention
        (attention_scores) -> Pattern [batch num_heads query key]: The attention scores entering the softmax, masked
        (attention_premix) -> HeadOutputs [batch seq num_heads head_dim]: Every head's output side by side, as the output projection takes it
        (attention_output) -> Residual [batch seq hidden_size]: What the attention adds to the residual stream
        (attention_probabilities) -> Pattern [batch num_heads query key]: The attention pattern the values are mixed with
        (attention_head_outputs) -> HeadOutputs [batch seq num_heads head_dim]: The per-head outputs before the output projection
      )
      (mlp): YoutuMLP(
        (gate_proj): Linear(in_features=2048, out_features=6144, bias=False)
        (up_proj): Linear(in_features=2048, out_features=6144, bias=False)
        (down_proj): Linear(in_features=6144, out_features=2048, bias=False)
        (act_fn): SiLUActivation()
        (mlp_activation) -> Neurons [batch seq intermediate_size]: The activation function's output
        (mlp_neurons) -> Neurons [batch seq intermediate_size]: What the down projection receives: act(gate)·up on a gated MLP, the activation itself otherwise
        (mlp_output) -> Residual [batch seq hidden_size]: What the MLP adds to the residual stream
      )
      (input_layernorm): YoutuRMSNorm((2048,), eps=1e-06)
      (post_attention_layernorm): YoutuRMSNorm((2048,), eps=1e-06)
      (layer_input) -> Residual [batch seq hidden_size]: The residual stream entering the block
      (attention_norm_output) -> Residual [batch seq hidden_size]: The normed stream entering the attention
      (layer_mid) -> Residual [batch seq hidden_size]: The residual stream after the attention add
      (mlp_norm_output) -> Residual [batch seq hidden_size]: The normed stream entering the MLP
      (mlp_input) -> Residual [batch seq hidden_size]: What the MLP is called with
      (layer_output) -> Residual [batch seq hidden_size]: The residual stream leaving the block, a tensor even when the block returns a tuple
    )
  )
  (norm): YoutuRMSNorm((2048,), eps=1e-06)
  (logits) -> Logits [batch seq vocab_size]: The logits the model returns, after anything it does past lm_head (a softcap, a scale)
  (token_embeddings) -> Residual [batch seq hidden_size]: The embedding module's output; layers[0].input is what enters the first block
  (final_norm_output) -> Residual [batch seq hidden_size]: The residual stream after the final norm, what lm_head reads
  (lm_head_output) -> Logits [batch seq vocab_size]: The head's logits before anything the model does past lm_head (a softcap, a scale)
  (next_token_probs) -> NextTokenProbs [batch vocab_size]: The next-token distribution at the last position; derived, read-only
  (input_ids) -> Tokens [batch seq]: The token ids the model was called with
  (attention_mask) -> Tokens [batch seq]: The attention mask the model was called with; zeros are padding
  (input_size): [batch, seq] of the current call; read-only
)

Config

model.num_layers32
model.hidden_size2048
model.num_heads16
model.num_kv_heads16
model.head_dim128
model.qk_head_dim192
model.vocab_size128256
model.intermediate_size6144
architecturesYoutuForCausalLM
num_hidden_layers32
hidden_size2048
intermediate_size6144
num_attention_heads16
num_key_value_heads16
head_dim64
vocab_size128256
max_position_embeddings131072
hidden_actsilu
rms_norm_eps1e-06
tie_word_embeddingsTrue
dtypebfloat16
q_lora_rank1536
kv_lora_rank512
qk_nope_head_dim128
qk_rope_head_dim64
v_head_dim128
attention_biasFalse

The Notes

03 · what to know before you run

The block, in order

h   = x + self_attn(input_layernorm(x))      # multi-head latent attention
out = h + mlp(post_attention_layernorm(h))   # a dense gated MLP on every block

Llama's pre-norm block and Llama's names. The block adds each sublayer's output to the stream and returns a tensor, so the identity is the plain sum:

with model.trace(prompt):
    x = model.layers[1].input.save()
    attn = model.layers[1].self_attn.attention_output.save()
    mlp = model.layers[1].mlp.mlp_output.save()
    out = model.layers[1].layer_output.save()

torch.testing.assert_close(x + attn + mlp, out)

Latent attention: queries and keys are wider than values

attention_queries and attention_keys are qk_head_dim wide (qk_nope_head_dim + qk_rope_head_dim, 128 + 64 = 192), attention_values and attention_head_outputs head_dim (v_head_dim, 128). transformers sets the config's head_dim to qk_rope_head_dim (64), which no served value has. Queries go through a compressed rank (q_lora_rank 1536: q_a_proj, q_a_layernorm and q_b_proj, with no q_proj module). The scores are scaled by qk_head_dim ** -0.5 (self_attn.scaling, 0.0722): the checkpoints have no YaRN scaling.

Keys and values are expanded from one latent per token (kv_lora_rank 512) for every head, so attention_keys has num_heads (16) heads, and the last 64 dimensions of every head's key are one shared rotary key. The attention interior needs attn_implementation="eager".

attn = model.layers[1].self_attn
with model.trace(prompt):
    q = attn.attention_queries.save()          # [1, 16, seq, 192]
    k = attn.attention_keys.save()             # [1, 16, seq, 192]
    v = attn.attention_values.save()           # [1, 16, seq, 128]

assert torch.equal(k[:, 0, :, -64:], k[:, 1, :, -64:])   # one rotary key

An edit to one head's last 64 key dimensions reaches that head only: the expansion has already copied the rotary key into every head when attention_keys is read.

The rotary part is served de-interleaved

rope_interleave is true: the attention rotates pair (2i, 2i + 1) of the projection into dimensions i and i + 32 of the rotary part. attention_queries[..., -64:] and attention_keys[..., -64:] are in that rotate-half order, not the order of q_b_proj's output: at position 0, where rotary is the identity, the served rotary part is the projection's even dimensions followed by its odd ones.

with model.trace(prompt):
    proj = attn.q_b_proj.output.save()
with model.trace(prompt):
    q = attn.attention_queries.save()

rope = proj.view(1, -1, model.num_heads, model.qk_head_dim)[0, 0, 0, -64:]
torch.testing.assert_close(q[0, 0, 0, -64:], torch.cat([rope[0::2], rope[1::2]]))

The readout is tied

lm_head.weight is embed_tokens.weight, so a direction read off the unembedding is also the embedding's: a write to one weight changes both. logits is lm_head.output, with no softcap or scale, and project_on_vocab applies norm and then the tied head.

From the family module

04 · nnterp/families/youtu.py

Youtu-LLM (YoutuForCausalLM).

Llama's tree and Llama's block: model.{embed_tokens, layers[i].{input_layernorm, self_attn, post_attention_layernorm, mlp}, norm} and lm_head, the residual added in the block, which returns a bare tensor. Attention is DeepSeek-V2's multi-head latent attention through the shared eager forward: queries and keys are qk_head_dim wide (qk_nope_head_dim + qk_rope_head_dim), values v_head_dim, and the interface sees num_heads key/value heads. The config maps head_dim to qk_rope_head_dim, the rotary part only, so both widths come from deepseek_v2's size functions. The MLP is dense on every block.